About the role
The Perception team at Zoox creates the "eyes and ears" of our self-driving robots. Navigating safely and efficiently in complex environments requires detecting, classifying, tracking, and understanding various attributes of surrounding objects—all in real-time and with exceptional accuracy.
As an engineer in the Scene Understanding team, you will develop advanced Vision-Language-Action (VLA) models that perceive our vehicle's surroundings to identify hazards and make driving suggestions. You will utilize VLA models for detecting rare events and ensuring safe driving in these situations. You'll work with state-of-the-art machine learning models that operate in real-time on our robotaxi platform with minimal latency. Collaborating with world-class engineers and researchers across sensors, planning, and other teams, you'll have access to premium sensor data and cutting-edge infrastructure to validate your algorithms in real-world conditions.
In this role, you will...
Design and train Vision-Language-Action (VLA) solutions for robotaxis
Lead end-to-end data strategy, including mining, auto-labeling, and dataset construction to power our ML flywheel
Lead the full post-training stack for VLMs and VLAs, including Continual Pre-training (CPT) on domain-specific driving data, Supervised Fine-Tuning (SFT) for instruction following.
Utilize our large-scale data pipelines and ML infrastructure to research, prototype, and deploy solutions that improve driving behavior
Partner with cross-functional teams to integrate perception signals
Qualifications
MS or PhD in Computer Science or related field
Background in deep learning solutions for VLM and VLA models
Track record in post-training large-scale models, CPT, SFT, RL
Hands-on experience with production ML pipelines, including dataset creation, training frameworks, and metrics
Expertise in Python libraries (PyTorch, NumPy, Pandas, VLLM)
Bonus Qualifications
Deep knowledge of cutting-edge computer vision techniques
Publications in top-tier conferences (CVPR, ICCV, RSS, ICRA)
Experience with integrating large language models to various tasks.